Will computers ever solve chess?

...fools' thoughts ...to solve chess.

Weak players always want to solve chess, they envy stronger players.

I think not, even with the advent of quantum computing - simply not enough power. however, an opening could probably be solved if they devoted a lot of hardware and eliminated transpositions, as well as stopping when a theoretical endgame position is reached, and not bothering to study lines in which the evaluation goes below, say, plus or minus three.

How can you be so sure that white's advantage is winning? Maybe with best play white will end up a minor piece up, and that is not enough to score a win. 

Because nobody understands what you said. Here is what I imagine if the players coorporate: Chess is a solved mate in 2 for black

I know xman's post is meant in jest, but perhaps that example can enlighten some who think chess is already solved or can be solved by humans.  1.f4 e5 2.g4 has been solved. 1.f3 e6 2.g4 has also been solved. 1.g4 e5 2.f4 has been solved, too. But that doesn't mean all of chess has been solved. 

Solved:

1.f4 e5 2.g4

1.g4 e5 2.f4

1.f3 e6 2.g4

1.g4 e6 2.f3

Also, any number of 3-move mates (and longer ones with meaningless N moves thrown in) have also been solved:

For example, 1.g3 e6 2.f3 e5 3.g4

Such a simple two-move mate sheds some light on how complex the problem is. 

______________

Chess players know that chess is a draw with best play, we just haven't gone through every variation to demonstrate it yet.

Chess players have been analysing chess for centuries, trying to come up with the best moves to force a win. As of today, there is no single best line in chess, no single line that gives White a clear edge against best play. Chess has far too large a draw margin for the game to be anything but a draw from a theoretical perspective.

We haven't gone through every single variation yet, but there are many ways of knowing. Science and philosophy agree that there are many ways of knowing, and that there isn't a bright line between belief and knowledge. Scientists accept something as true if there is a less than 5 sigma chance of it being false. By that standard, chess is a draw. There's less than 1/3.5million chance of this being wrong. We have 10s of millions of games have been played. Not a single game has been won without a mistake being made. 

Long before computers created tablebases, chess players knew that you can't force a mate with 2 Ns vs a lone King. This is provable not by going through every possible position, but by analysing a few key positions such as the one below.

You don't need to analyse every single K+2N vs K position to know that you can't force a win. 

Good point, SmyslovFan. Do you think identifying more of this kind of generalization rule can make major inroads into the brute-force task, or only nibble at the edges?

Today, nibbling at the edges is a pretty huge chunk. We can correctly evaluate almost every position where there are 7 pieces or fewer, and engines can correctly work out many (most?) positions with slightly more pieces by accessing that information. 

So we've already made major inroads, and will make more.

I am glad that SmyslovFan is also saying what I have pointed out multiple times in this thread, as I was feeling like no one understood.

You do not need enumeration to solve a game.

In fact, as I pointed out long ago, even infinite games can have solutions.  The example I give is the one where each side has a single piece on an infinite strip one square wide (1 x oo), where each piece can move 1 square and take if the other piece is on it.  This is easy to solve - optimal strategy is for the player that is odd squares away on their move to move towards the other player and for the player even squares away on their move to move away.  We solve this by simple mathematical invariants, despite there being an infinite number of possible positions.

And in fact, as both I and SmyslovFan have pointed out, many endgames were originally solved with these techniques.  And in recent times, these techniques have been extended to certain piece combination endgames using some really interesting mathematical techniques, mostly from places like linear algebra and related.

The point is, it may be possible to find a mathematical function that can evaluate a position and say with perfect play whether it is winning for white, black, or a draw, just using the relationships of the pieces and no enumeration.  And if such a function is found, evaluation on the original position will tell you who wins the game with perfect play, and to actually find a move that makes it, one simply needs to evaluate that function on all the 1-move alternatives.

Obviously, we don't have anything like that right now.  But that is a perfectly valid possibility on how chess might be "solved" without needing all the atoms in the universe to store positions, and it is a common technique in modern game theory.

Both of you are correct in general: there do exist games where solutions exist that circumvent the game complexity.

Both of you are also wrong in the case of chess, I am sure. There are no comparable games where this is possible, and there is no reason to believe chess is one such.

The rules of chess are sufficiently arbitrary as to not allow certainty by any major shortcuts. Let me make it clear that what would be required would be a way of classifying superficially balanced positions as DEFINITELY drawn or DEFINITELY lost WITHOUT WORKING OUT WHAT MIGHT HAPPEN. 

The fact that this is utterly implausible is the reason that the most absurd and amusing chess problems exist. The best that is feasible is to achieve something analogous to the successful weak solving of checkers. The difference is that in the case of checkers this involved a game complexity of around 10^20 compared to 10^47 in the case of chess.

Roughly speaking, a factor of 10^27 greater computing power is needed.

Note that if a weak solution for both players is achieved, this solves the game for practical purposes (in the case of chess, this would require a forced drawing strategy for each player, barring the very unlikely existance of a forced winning strategy for one of them).

What is far more feasible is to achieve an approximate solution that is so close to entirely adequate that the difference is indetectible. This is something like probabilistic primality testing. Rigorous testing of primality of a number can be very computationally intensive (although modern methods now achieve quite low powers of log(n))  but there are methods that work a very high percentage of the time that are much faster. If such a method is successful often enough that you would be unlucky to ever be wrong, it would be almost as useful as a deterministic method, but would not be solving the primality question.

correct, you are. Ignorant people like to talk about tree without knowing its name.

From a recent film I believe the tree was called Groot

its so funny how some think that the game of chess can be solved lol solve how? the number of possible positions is greater than the number of atoms in the visible universe and yet we have "experts" who claim it is possible lol

good luck solving it !! with trees and vegetables lool

Guys, the problem is not with the leaves of the tree. The problem is that it isn't a shrub. 

why tree and not a potato field ;)

I agree with smylov fan and I like that primality testing was brought up in Elroch's post.

There are hundreds, even thousands of methods to test primes. There are at least dozens of prime test methods which are completely fool proof. They are never wrong. 

In addition, these methods can prove a number is composite; if a number fails a perfect prime number test, the the number is composite.

However, just because we absoutely with 100% certainty proved a number is composite does not mean we know its factors.

We can take the largest prime number currently known and add 1, and that will definitely be composite. That is solved.

But what are its factors? We have no idea. We have no way to even find out. The largest prime number (+1 added) is over 22 million digits long, computers cannot begin to brute force its factors.

In the same way, it's concievable to be able to solve chess without working through 10^120 possible variations. Someone who is clever can determine just by aspects of the position and an arbitriarly large, but doable amount of computer power that a position is won or lost or drawn in n moves. 

Here is a simple example:

Does white win?

Yes, white wins.

Did you have to calculate an infinite amount of games to figure that out?

White has 9 places he can move his king, and black does too. Then they have 9 places from there, and 9 places from there, and 9 places from there...

To account for reaching edges, and other stuff let's assume that there are an average of 50 legal king moves per turn.

Did you have to calculate through 2500 variations of the kings running around to conclude that white's pawn has to move? Then did you move the pawn forward 1, and go through the 2500 variations again? How do you know white wins? Did you try everything?

Of course you didn't try everything. You don't need to, and neither does a computer.